Plenary Talk 1
Presenter:
Prof. Tetsuro Ogi
Graduate School of System Design and Management
Keio University
Title:
Avatar Psychology for Communication in Networked Virtual and Augmented Reality Worlds
Abstract:
Currently, avatars are used in communication in various situations, such as in the metaverse society or robot avatars in the real world. However, there has not been much discussion about the psychological issues of users when communicating using avatars, such as how they feel when operating an avatar, how they feel about the other user’s avatar, or how their feelings change depending on the avatar’s appearance. Though psychology has developed as a academic field to build a better society for humans in the real world, it is also necessary to build a academic research field of avatar psychology in the avatar society. In this presentation, the psychological issues in avatar communication will be discussed while introducing our recent research projects.
Biography:
Tetsuro Ogi obtained a master’s degree in Mechanical Engineering from the University of Tokyo in 1986, and joined Mitsubishi Research Institute as a researcher. He obtained a Ph.D. in Mechanical Engineering from the University of Tokyo in 1994 and was named associate professor at Intelligent Modeling Laboratory, the University of Tokyo in 1996. In 2004, he moved to the Department of Computer Science, Graduate School of Systems and Information Engineering, University of Tsukuba as associate professor, and he is a professor at Graduate School of System Design and Management, Keio University from 2008. His research interests include human interface, virtual reality, high presence communication, etc.
Plenary Talk 2
Presenter:
Prof. Ryo Yoshida
Director (Professor), Research Center for Materials Informatics, The Institute of Statistical Mathematics (ISM), Research Organization of Information and Systems (ROIS)
Team Leader, Polymer Omics Team, Advanced General Intelligence for Science Program (AGIS), RIKEN-TRIP
Title:
Omics-scale simulation database transferable to real-world artificial intelligence applications
Abstract:
Creating large-scale foundational datasets represents the first milestone for enabling scientific innovation driven by artificial intelligence (AI). In many scientific disciplines—except for a few AI-advanced fields like computer vision and natural language processing— the development of open, AI-ready foundational databases has lagged significantly behind. To overcome this challenge, high-throughput simulations for generating massive computational datasets and Sim2Real machine learning to bridge the gap between the computational and real worlds play a crucial role.
To tackle the data scarcity in materials science, we developed the world’s largest computational database for polymer materials, comprising diverse physical properties for more than 100,000 polymer species with a wide range of higher-order structures and applied external fields. This omics-scale database has been built using RadonPy, a Python-based pipelining software for fully automating all-atom molecular dynamic simulations for polymeric materials [1]. Machine learning models pretrained on the RadonPy database can be readily fine-tuned to a diverse array of real-world downstream tasks using limited experimental data to achieve outstanding generalization in the real-world prediction tasks—beyond the capabilities of models trained from scratch. Remarkably, it is shown that the generalization capability of these transferred models improves significantly as the size of the RadonPy database increases, following a power-law scaling across a wide range of real-world prediction tasks [2]. This unprecedented omics-scale simulation database has uncovered vast, previously unexplored materials spaces that remain beyond the reach of current experimental methodologies.
References:
[1] Hayashi, Y., Shiomi, J., Morikawa, J., Yoshdia, R., RadonPy: automated physical property calculation using all-atom classical molecular dynamics simulations for polymer informatics. npj Comput Mater 8, 222 (2022). https://doi.org/10.1038/s41524-022-00906-4
[2] Minami, S., Hayashi, Y., Wu, S., Fukumizu, K., Sugisawa, H., Ishii, M., Kuwajima, I., Shiratori, K., Yoshida, R., Scaling law of Sim2Real transfer learning in expanding computational materials databases for real-world predictions. npj Comput Mater 11, 146 (2025). https://doi.org/10.1038/s41524-025-01606-5
Biography:
Ph.D. from The Graduate University for Advanced Studies (SOKENDAI) in 2004. Ryo Yoshida is a Professor at the Institute of Statistical Mathematics (ISM), Research Organization of Information and Systems, and serves as the Director of Research Center for Materials Informatics. Since 2025, he has been the vice director at the ISM. As a machine learning researcher, he has made significant contributions to creating the scientific foundation of data-driven materials research. In recent years, he is leading an industry-academia consortium comprising over 260 researchers and played a central role in the joint development of RadonPy—a software platform for fully automated computational experiments of polymer materials—and the world’s largest polymer materials database. Furthermore, in 2025, he launched the “Polymer Omics” team at RIKEN’s TRIP-AGIS program, where he is advancing the development of a fully autonomous materials discovery system that integrates AI, automated simulations, and automated experiments.
Plenary Talk 3
Presenter:
Prof. Sadayoshi Murakami
Department of Nuclear Engineering
Kyoto University
Title:
Real-Time Predictive Control for Fusion Plasma Using Data Assimilation
Abstract:
Since the fusion reaction rate depends strongly on plasma temperature and density, the future reactor will require precise control to start up and maintain a stable burning plasma. On the other hand, due to the intense neutron radiation, only limited measurement data can be obtained in fusion reactors. For this reason, many efforts have been made to develop an integrated simulation code that accurately predicts plasma temperature and density. The complexity of integrated simulation, where each model contains uncertainties, is a significant challenge in obtaining reliable results. In addition, uncertainties not included in the model, such as wall conditions, may affect the plasma performance. It isn’t easy to get fast and accurate predictions using the present integrated simulation code.
To overcome these problems, we develop an adaptive predictive control system for fusion plasmas based on data assimilation techniques, which integrates a predictive model (digital twin) adaptation using real-time measurements and control estimation robust against model and observation uncertainties. The main part of the control system, ASTI, predicts the probability distribution of future plasma states and estimates the optimal control input and the actual plasma state based on Bayes’ theorem.
In this study, the ASTI-centered control system has been implemented in the Large Helical Device (LHD). We employ the integrated simulation code, TASK3D, as the digital twin of the LHD plasma in ASTI. The ASTI system successfully applied to control the electron and ion temperatures and electron density. The control experiments demonstrate the effectiveness of the data assimilation-based control approach, which allows the synergistic interaction of measurement, heating, fueling, and simulation. This approach can provide a flexible platform for digital twin control of future fusion reactors.
Biography:
Dr. Sadayoshi Murakami is currently a Professor in the Department of Nuclear Engineering, Kyoto University. He received his Ph.D. in 1992 from Hiroshima University. After he received his Ph.D., he worked at the National Institute for Fusion Science from 1992 to 2002. He has been working at Kyoto University since 2003. He has been conducting research on plasma heating and confinement in stellarators, which are three-dimensional magnetic field configurations, intending to realize a magnetic confinement fusion reactor. In particular, he has analyzed the physical mechanisms of heat and particle transport through neoclassical transport, which involves the transport of high-energy particles due to heating and diffusion processes caused by Coulomb collisions, using both theoretical and experimental approaches. Recently, he has been advancing research on real-time control of plasma using data assimilation techniques.